brianloss commented on a change in pull request #2123:
URL: https://github.com/apache/accumulo/pull/2123#discussion_r641692460
##########
File path:
test/src/main/java/org/apache/accumulo/test/manager/SuspendedTabletsIT.java
##########
@@ -91,21 +100,65 @@ public void configure(MiniAccumuloConfigImpl cfg,
Configuration fsConf) {
cfg.setProperty(Property.TABLE_SUSPEND_DURATION, SUSPEND_DURATION + "s");
cfg.setClientProperty(ClientProperty.INSTANCE_ZOOKEEPERS_TIMEOUT, "5s");
cfg.setProperty(Property.INSTANCE_ZK_TIMEOUT, "5s");
- cfg.setNumTservers(TSERVERS);
+ // Start with 1 tserver, we'll increase that later
+ cfg.setNumTservers(1);
// config custom balancer to keep all metadata on one server
- cfg.setProperty(HostRegexTableLoadBalancer.HOST_BALANCER_PREFIX +
MetadataTable.NAME, "*");
- cfg.setProperty(HostRegexTableLoadBalancer.HOST_BALANCER_OOB_CHECK_KEY,
"7s");
- cfg.setProperty(Property.TABLE_LOAD_BALANCER.getKey(),
- HostRegexTableLoadBalancer.class.getName());
+ cfg.setProperty(HostRegexTableLoadBalancer.HOST_BALANCER_OOB_CHECK_KEY,
"1ms");
+ cfg.setProperty(Property.MANAGER_TABLET_BALANCER.getKey(),
+ HostAndPortRegexTableLoadBalancer.class.getName());
+ }
+
+ @Before
+ public void setUp() throws Exception {
+ super.setUp();
+
+ try (AccumuloClient client =
Accumulo.newClient().from(getClientProperties()).build()) {
+ // Wait for all tablet servers to come online and then choose the first
server in the list.
+ // Update the balancer configuration to assign all metadata tablets to
that server (and
+ // everything else to other servers).
+ InstanceOperations iops = client.instanceOperations();
+ List<String> tservers = iops.getTabletServers();
+ while (tservers == null || tservers.size() < 1) {
+ Thread.sleep(1000L);
+ tservers = client.instanceOperations().getTabletServers();
+ }
+ HostAndPort metadataServer = HostAndPort.fromString(tservers.get(0));
+ log.info("Configuring balancer to assign all metadata tablets to {}",
metadataServer);
+ iops.setProperty(HostRegexTableLoadBalancer.HOST_BALANCER_PREFIX +
MetadataTable.NAME,
+ metadataServer.toString());
+
+ // Wait for the balancer to assign all metadata tablets to the chosen
server.
+ ClientContext ctx = (ClientContext) client;
+ TabletLocations tl = TabletLocations.retrieve(ctx, MetadataTable.NAME,
RootTable.NAME);
+ while (tl.hosted.keySet().size() != 1 ||
!tl.hosted.containsKey(metadataServer)) {
+ log.info("Metadata tablets are not hosted on the correct server.
Waiting for balancer...");
+ Thread.sleep(1000L);
+ tl = TabletLocations.retrieve(ctx, MetadataTable.NAME, RootTable.NAME);
+ }
+ log.info("Metadata tablets are now hosted on {}", metadataServer);
+ }
+
+ // Since we started only a single tablet server, we know it's the one
hosting the
+ // metadata table. Save its process reference off so we can exclude it
later when
+ // killing tablet servers.
+ Collection<ProcessReference> procs =
getCluster().getProcesses().get(ServerType.TABLET_SERVER);
+ assertEquals("Expected a single tserver process", 1, procs.size());
+ metadataTserverProcess = procs.iterator().next();
+
+ // Update the number of tservers and start the new tservers.
+ getCluster().getConfig().setNumTservers(TSERVERS);
+ getCluster().start();
}
@Test
public void crashAndResumeTserver() throws Exception {
// Run the test body. When we get to the point where we need a tserver to
go away, get rid of it
// via crashing
suspensionTestBody((ctx, locs, count) -> {
- List<ProcessReference> procs =
- new
ArrayList<>(getCluster().getProcesses().get(ServerType.TABLET_SERVER));
+ // Exclude the tablet server hosting the metadata table from the list
and only
+ // kill tablet servers that are not hosting the metadata table.
+ List<ProcessReference> procs =
getCluster().getProcesses().get(ServerType.TABLET_SERVER)
+ .stream().filter(p ->
!metadataTserverProcess.equals(p)).collect(Collectors.toList());
Review comment:
Done
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
